[XLA:CPU] Optimize StackFrames Proto index access and Dynamically Cap MultiModuleDriver parallel compilation concurrency#44243
Open
copybara-service[bot] wants to merge 1 commit into
Open
Conversation
copybara-service
Bot
force-pushed
the
test_931295728
branch
2 times, most recently
from
51ce453 to
c1d4a6f
Compare
… MultiModuleDriver parallel compilation concurrency - Rewrote `StackFrames::IsPrefix` to traverse stack frames directly via `.parent_frame_id()` protobuf indexes rather than allocating full `HloStackFrame` structural copies on every hop, resolving extreme compile-time O(N) memory allocation hotspots during call graph metadata propagation. - Capped `MultiModuleDriver::Compile` parallel submodule compilation to dynamically match the dimensions of `CompileOptions::thread_pool` (or a safe fallback limit of 8 concurrent LLVM compilations) to prevent Out-Of-Memory (OOM) hard freezes and virtual memory thrashing when compiling massively split models (like `torax`) under `FAST_COMPILE`. PiperOrigin-RevId: 931295728
copybara-service
Bot
force-pushed
the
test_931295728
branch
from
c1d4a6f to
98b170b
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[XLA:CPU] Optimize StackFrames Proto index access and Dynamically Cap MultiModuleDriver parallel compilation concurrency
StackFrames::IsPrefixto traverse stack frames directly via.parent_frame_id()protobuf indexes rather than allocating fullHloStackFramestructural copies on every hop, resolving extreme compile-time O(N) memory allocation hotspots during call graph metadata propagation.MultiModuleDriver::Compileparallel submodule compilation to dynamically match the dimensions ofCompileOptions::thread_pool(or a safe fallback limit of 8 concurrent LLVM compilations) to prevent Out-Of-Memory (OOM) hard freezes and virtual memory thrashing when compiling massively split models (liketorax) underFAST_COMPILE.